Fal, which runs AI image and video models for Adobe and Canva, says it raised $1.65 billion
Unlike data center companies that rent out chips by the hour, Fal mostly charges for each picture or video clip its models make.
Fal, a startup that runs AI image, video and audio models for other companies, said Tuesday it has raised $1.65 billion across two funding rounds.
Sequoia Capital, the Menlo Park venture firm, led a Series E earlier this year. Bessemer Venture Partners led the new Series F, with chipmaker Nvidia as a strategic investor, Fal said. A Series F is a company's sixth major round of outside investment. Fal did not disclose its valuation.
What Fal does
Fal sells inference, the work of running a trained AI model each time someone asks it to make a picture or clip. Developers reach more than 1,000 ready-to-use models, including Fal's own tuned versions, through one API, a connection programs use to talk to each other, according to Fal's documentation. Bessemer said Adobe, Canva, Shopify and Amazon have moved production work to Fal.
How it differs from data center companies
Cloud and data center companies such as CoreWeave mainly rent out computer time. CoreWeave's price list, for example, charges $49.24 an hour on demand for a machine with eight Nvidia H100 chips. Customers then have to install, tune and run the AI models themselves.
Fal mostly sells the finished result instead. On its main service, a customer picks a model, sends a request and pays per output, such as 2.5 cents per image from the Flux dev model or a set price per second of video, according to Fal's pricing page. Fal says customers are not charged for server errors or time spent waiting in line.
Fal also rents raw computing for customers who want to run their own models. Its Serverless service bills by the second of chip time and can grow from zero to thousands of chips as demand rises, Fal says, and its Compute service rents dedicated H100 machines by the hour.
Bessemer said image and video models need different software than text chatbots, because they refine a whole frame over many passes, and that Fal tuned its system for them "down to the kernel," the lowest level of chip code. Bessemer also described Fal as a two-sided marketplace: developers get one place to try thousands of models, and AI labs such as Black Forest Labs use it to reach customers. Fal did not say whether it owns its own data centers.
Fal said the money will pay for more computing capacity and research in 2027. Fal was founded by Burkay Gur and Gorkem Yurtseven.
Source: Fal blog, Fal post on X and Fal documentation and pricing pages, Nov. 3, 2026; Bessemer Venture Partners, "Investing in fal," Nov. 3, 2026; CoreWeave pricing page